Research On 2D Image-Based Unsupervised 3D Model Retrieval Method

With the development of 3D modeling technology and its wide application in different fields,the number of 3D models increases rapidly,making 3D model retrieval a hot topic in current research.Compared with other 3D model retrieval methods,2D image-based unsupervised 3D model retrieval takes the 2D images which have rich labels and are easy to obtain as the queries,and also takes into account the difficulties of labeling 3D models.Therefore,it has high practical value and has attracted extensive attention in the related field.2D image-based unsupervised 3D model retrieval is a retrieval task involving crossdomain adaptation problem,which is different from general domain adaptation problem or single domain retrieval task.There are two main challenges in this task:excessive domain gap and weak feature expression in semantics.In view of the above two challenges,this thesis proposes a cross-domain 3D model retrieval method based on disentangled feature learning,and further proposes a progressive cross-domain 3D model retrieval method based on memory mechanism and a cross-domain 3D model retrieval method based on semantic enhancement.Firstly,the designed disentangled feature learning enables to disentangle the twisted raw features into the isolated domain-invariant features and domain-specific features,where the following alignment is forced to focus on the former to narrow the domain gap.On this basis,the memory mechanism can gradually improve the adaptability of the model to the very different target domains.The semantic enhancement brings class-level feature alignment to the global feature alignment,and mines semantic consistency in unsupervised target domain to enhance the semantic discrimination in features,making features more suitable for the retrieval task.A large number of experiments are conducted on the public datasets MI3DOR and MI3DOR-2 to verify the feasibility of the proposed methods,and the results also demonstrate their superiority.